A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 11, 23:45 UTC
Do not guard during noop elimination pass (#30095)
Signed-off-by: Laith Sakka <lsakka@meta.com>
Laith Sakka committed
Dec 4, 2025 at 20:10 UTC
5867819eaffe3c939c0920c15d5048cb7f9129f8
1 file changed
+2
-6
vllm/compilation/noop_elimination.py
+2
-6
@@ -5,6 +5,7 @@ from collections.abc import Iterable
5
6
import torch.fx
7
from torch import SymInt
8
+from torch.fx.experimental.symbolic_shapes import statically_known_true
9
10
from vllm.logger import init_logger
11
@@ -116,12 +117,7 @@ class NoOpEliminationPass(VllmInductorPass):
117
2. The dimensions both correspond to the same SymInt
118
"""
119
# Case 1
119
- if isinstance(i_dim, int) and isinstance(dim, int):
120
- return dim == i_dim
121
- # Case 2
122
- if isinstance(i_dim, SymInt) and isinstance(dim, SymInt):
123
- return dim == i_dim
124
- return False
120
+ return statically_known_true(dim == i_dim)
121
122
def all_dims_equivalent(
123
self, dims: Iterable[int | SymInt], i_dims: Iterable[int | SymInt]